[SOLVED] Proxmox VE unstable with 4 RAM modules

[gmjs]

Hello,

Does anyone have any experience of running Proxmox VE 6 on an AMD Ryzen 5 with four RAM modules?

I have purchased components and have tested the following in the server:

Config 1.
---
AMD Ryzen 5 1600 processor
Gigabite AB320M Gaming 3 motherboard
4x 16GB DDR4 Ballistix RAM @ 2400MHz (JEDEC, stock)
(With this configuration, the RAM *is* listed on Motherboard's QVL.)


Config 2.
---
AMD Ryzen 5 1600 processor
Gigabite AB320M Gaming 3 motherboard
4x 16GB DDR4 Patriot RAM @ 2400MHz (JEDEC, stock)
(With this configuration, the RAM *is not* listed on Motherboard's QVL.)

Config 3.
---
AMD Ryzen 5 1600 processor
MSI B450M Pro-VDH MAX motherboard
4x 16GB DDR4 Patriot RAM @ 2400MHz (JEDEC, stock)
(With this configuration, the RAM *is* listed on Motherboard's QVL.)

Config 4.
---
AMD Ryzen 5 3600 processor
MSI B450M Pro-VDH MAX motherboard
4x 16GB DDR4 Patriot RAM @ 2400MHz (JEDEC, stock)
(With this configuration, the RAM *is* listed on Motherboard's QVL.)


and all configurations see kernel errors after anything from 1 day to 8 days' use.

With only two RAM modules (for 32GB RAM in total) the server runs without error.


I've run out of things to try to run this server with all four memory modules.


MemTest+86 reports faults after only a few seconds when I run it just after the server has crashed, but no errors after the NVRAM is cleared after a crash.


Has anyone else seen this problem? Is there anything I can do to resolve it?

I'd be very grateful for any suggestions---I've run out of my own ideas (and money ).

Many thanks.

 

[Alwin]

Does anyone have any experience of running Proxmox VE 6 on an AMD Ryzen 5 with four RAM modules?

Sounds like BIOS updates are needed. :/

 

[gmjs]

Hi Alwin, and thanks for the reply.

I updated the firmware on both boards to try to solve the error, but both still struggle with 4 DIMMs at stock speed.

I agree that it feels like a hardware issue---the errors are too random. I've had one VM out of 10 shutdown a couple of times, an SQL database become corrupt, a Tomcat application crash, and (most often) kernel panics on the host. Seemingly only when more than 3/4 of the total RAM (when using 4 modules) is in use. I also disable C-States.

Examples of errors I get:

Jun 22 18:14:56 vmhost kernel: [28645.910801] general protection fault: 0000 [#1] SMP NOPTI
Jun 23 19:45:50 vmhost kernel: [88922.005454] general protection fault: 0000 [#1] SMP NOPTI
Jul  6 09:12:53 vmhost kernel: [836447.007980] invalid opcode: 0000 [#1] SMP NOPTI
Jul 19 05:28:25 vmhost kernel: [36814.167540] BUG: Bad page state in process kvm  pfn:4c8c72

I've emailed AMD to see if they have any suggestions.

I hope it isn't an issue with the Proxmox VE kernel (I really like Proxmox VE---and with 2 DIMMs in the machine, it works beautifully).

Thanks.

 

[gmjs]

I've happened upon an interesting wiki site that suggests that 4 DIMMs are only supported at a speed of 1866MHz on the Ryzen 5 processors:

https://en.wikichip.org/wiki/amd/ryzen_5/1600

I'll be testing this when I can have some downtime!

 

[gmjs]

My server crashed even more quickly when running the memory at a slower speed.

I've given up and sent two modules back. Nothing I have tried will allow me to use 4 modules at once, and I've had no response from AMD.

 

[Alwin]

Try to disable all S-/P-states and see if that stabilizes the system.

 

[gmjs]

I've upgraded to Proxmox VE 7.0, and believe the following to be the case.

Support for AMD Ryzen processors (in my experience with 4 DIMMS in use) was terrible in the Linux Kernel version 5.4. I've read other posts that mention stability problems with AMD processors in Linux distributions that adopted the 5.4 release too.

I've had no problems at all with Linux Kernel version 4.19 and 5.10. It's just unfortunate 5.4 was a long-term support release.

I'm running Proxmox VE 7.0, with 64GB of RAM across 4 memory slots (Linux kernel 5.11) without issue.